GPUBudgets

GPUBudgets refers to mechanisms for constraining and allocating GPU resources across workloads to meet performance, power, and quality-of-service goals. The concept encompasses per-application or per-task budgets and is used at hardware, driver, and application layers to manage trade-offs among rendering fidelity, frame times, and energy use.

Common budget types include frame time budgets (a maximum time per frame to maintain a target frame rate), energy or thermal budgets (limits on power draw or the thermal envelope), memory budgets (limits on VRAM usage and working set), and compute or bandwidth budgets (limits on shader instructions, texture fetches, or memory bandwidth). Latency budgets may also be set for GPU-CPU interactions and task queues.

Enforcement is typically achieved through GPU schedulers, dynamic voltage and frequency scaling (DVFS), and quality-reducing techniques such as resolution scaling, adaptive sampling, or level-of-detail culling. Budgets can be static, defined by developers or system designers, or dynamic, adjusted at run-time based on thermal readings, battery level, or workload priority.

Applications include mobile devices aiming to extend battery life, data-center GPUs enforcing fair share among tenants, browsers or runtime environments that limit WebGPU workloads, and game engines that enforce frame pacing and quality targets. Designers must balance measurement overhead, budget granularity, and user-perceived quality.

See also: power management, GPU scheduling, memory budgeting, frame pacing.